Low-latency ethernet: the ubiquitous datacenter interconnect

ABSTRACT

It is well known in the HPC community that latency is critical and that CLOS architectures are key to scalable, low-latency, high-bandwidth systems. The critical element in the further scalability of these systems is the emergence of a robust, multi-vendor ecosystem for low-latency interconnect components with a low cost. Ethernet has the cross-market appeal, and comprises a majority of HPC systems, but is not yet seen as a performance leader. This presentation will expose the audience to new capabilities in Ethernet that make it suitable for the highest performance systems. In particular, TCP/IP/Ethernet interconnect technology now has: 1) 10 Gbps data rate 2) Extremely low latency 3) Full protocol offload and acceleration 4) Scalable, non-blocking topologies (fat trees, etc.) 5) Congestion management enhancements 6) Aggressive Ethernet-ecosystem pricing The presentation will include simulation results to show interconnect and congestion management performance in an HPC cluster from an end-to-end system perspective.

INDEX TERMS

Primary Classification:
  C. Computer Systems Organization
  C.2 COMPUTER-COMMUNICATION NETWORKS
      C.2.1 Network Architecture and Design

          Nouns: Ethernet

Additional Classification:
  B. Hardware
  B.4 INPUT/OUTPUT AND DATA COMMUNICATIONS
      B.4.3 Interconnections (subsystems)

  C. Computer Systems Organization
  C.2 COMPUTER-COMMUNICATION NETWORKS
      C.2.1 Network Architecture and Design
          Subjects: Network topology
      C.2.3 Network Operations
          Subjects: Network management

General Terms:
Design, Management